How to Perform Numpy Broadcasting using Python using dynamic arrays?

Broadcasting refers to how NumPy handles arrays of different dimensions during arithmetic operations. The smaller array is "broadcast" across the larger array to ensure compatible shapes for element-wise operations. This vectorizes operations in C rather than Python, making calculations faster and more memory-efficient.

Broadcasting eliminates the need for data copies while enabling efficient algorithm implementations. However, it can sometimes lead to memory overhead if not used carefully.

Broadcasting Rules

NumPy follows specific rules when broadcasting arrays ?

• Arrays are aligned from the rightmost dimension
• Dimensions of size 1 can be stretched to match larger dimensions
• Missing dimensions are treated as size 1
• Arrays must be compatible along each dimension

Scalar and Array Broadcasting

The simplest form of broadcasting occurs between a scalar and an array ?

import numpy as np

# Create array from 0 to 7
input_array = np.arange(8)
print("Input array:", input_array)

# Add scalar to array (broadcasting)
result = input_array + 9
print("After adding 9:", result)

Input array: [0 1 2 3 4 5 6 7]
After adding 9: [ 9 10 11 12 13 14 15 16]

The scalar 9 is automatically broadcast to match the array's shape, effectively adding 9 to each element.

Compatible Array Broadcasting

Arrays with compatible dimensions can be broadcast together ?

import numpy as np

# Create 3x4 array
array1 = np.arange(12).reshape(3, 4)
print("Array 1 shape:", array1.shape)
print("Array 1:\n", array1)

# Create 3x1 array  
array2 = np.arange(3).reshape(3, 1)
print("Array 2 shape:", array2.shape)
print("Array 2:\n", array2)

# Broadcasting addition
result = array1 + array2
print("Sum result:\n", result)

Array 1 shape: (3, 4)
Array 1:
 [[ 0  1  2  3]
 [ 4  5  6  7]
 [ 8  9 10 11]]
Array 2 shape: (3, 1)
Array 2:
 [[0]
 [1]
 [2]]
Sum result:
 [[ 0  1  2  3]
 [ 5  6  7  8]
 [10 11 12 13]]

Array2's single column is broadcast across all four columns of Array1, adding corresponding row values.

Multidimensional Broadcasting

Linear arrays can be broadcast with multidimensional arrays ?

import numpy as np

# Create 5x3 multidimensional array
array1 = np.arange(15).reshape(5, 3)
print("Array 1 shape:", array1.shape)

# Create 1D array with 3 elements
array2 = np.arange(3)
print("Array 2 shape:", array2.shape)

# Broadcasting works - 1D array broadcasts across rows
result = array1 + array2
print("Broadcasting result:\n", result)

Array 1 shape: (5, 3)
Array 2 shape: (3,)
Broadcasting result:
 [[ 0  2  4]
 [ 3  5  7]
 [ 6  8 10]
 [ 9 11 13]
 [12 14 16]]

Complex Broadcasting Example

Higher-dimensional arrays can also be broadcast ?

import numpy as np

# Create 4D array: (6, 5, 4, 2)
array1 = np.arange(240).reshape(6, 5, 4, 2)
print("Array 1 shape:", array1.shape)

# Create 3D array: (5, 4, 1) 
array2 = np.arange(20).reshape(5, 4, 1)
print("Array 2 shape:", array2.shape)

# Broadcasting result maintains larger array's shape
result = array1 + array2
print("Result shape:", result.shape)

Array 1 shape: (6, 5, 4, 2)
Array 2 shape: (5, 4, 1)
Result shape: (6, 5, 4, 2)

Array2 is broadcast along multiple dimensions to match Array1's shape, creating efficient element-wise operations.

Broadcasting Rules Summary

Conclusion

NumPy broadcasting enables efficient element-wise operations between arrays of different shapes without data copying. Understanding broadcasting rules helps optimize memory usage and computational performance in data science applications.